Nvidia's $500B AI financing plan raises 2008-style risk concerns

Nvidia is assembling a $500 billion AI infrastructure financing consortium with six Wall Street firms, but Bloomberg macro strategist Simon White warns the financialization of GPU compute echoes the AIG-CDO structure that preceded the 2008 housing collapse.

"Modern compute has emerged as a scarce, mission-critical asset class," Apollo President Jim Zelter said. White, however, drew a direct parallel to AIG's insurance on CDO tranches, noting Nvidia will backstop up to 25 percent of GPU collateral value in some transactions.

The consortium includes Apollo, BlackRock's Global Infrastructure Partners, Blackstone, Brookfield, Goldman Sachs and KKR. Nvidia shares fell 3.1 percent in midday trading Monday despite the announcement. The deal follows a $100 billion Brookfield-Nvidia program and a $35 billion Apollo-Blackstone-Broadcom platform. Goldman Sachs projects global AI infrastructure investment will exceed $1 trillion in 2026 alone.

The financing structure treats AI compute as an investable asset class with usage-linked revenue, but White argues financialization is masking real supply-demand imbalances. GPU utilization averages just 5 percent, according to Cast AI, while ClearML found nearly half of enterprises waste millions on idle chips — data points that suggest the market may already be oversupplied.

The deal marks a fundamental shift in Nvidia's role. Once strictly a hardware supplier, the Santa Clara-based company now functions as a capital-mobilization engine, connecting institutional money to the physical infrastructure required to sustain the AI boom. The financing will support investments across the AI stack — from advanced chips and networking gear to the massive energy facilities needed to run them. Nvidia is also pursuing parallel arrangements outside the Wall Street consortium, including a $500 billion partnership with SK Group covering AI factories and memory technology.

The AIG Parallel

White's comparison centers on how financial engineering can distort risk perception. In the housing bubble, AIG sold insurance on senior CDO tranches, allowing banks to reduce capital requirements and distribute more toxic debt. When default correlations spiked, the system collapsed. Nvidia's structure is similar: by guaranteeing GPU resale values, the company makes debt backed by AI hardware appear safer than it is.

The SEC recently relaxed securitization rules for data center bonds, which White said could accelerate the process. He compared this to the moment pension funds entered the housing market — typically a sign that risk is spreading to the broader liability side of the financial system. The GPU-backed debt market has already grown explosively this year, and the new agreements push securitization and distribution deeper.

Efficiency Gains Undermine GPU Pricing

White's core concern is that financialization is systematically suppressing oversupply signals. The standard metric for GPU rental is cost per hour; for inference, it's cost per token. But what matters to end users is the output — how much intelligence value each dollar buys.

Intelligence costs are falling, but this isn't immediately reflected in GPU-level pricing. If a model can accomplish the same task with 10 times fewer tokens while GPU per-token pricing stays flat, each GPU's effective efficiency has increased tenfold. As outcome-based pricing gains traction among enterprise customers, model providers face pressure to reduce token consumption, which would push GPU rental prices lower.

The Jevons paradox — the idea that efficiency gains drive increased demand — doesn't hold indefinitely, White argued. Path dependency is the real risk: at current price levels, demand growth for intelligence will eventually slow enough to trigger a prolonged decline in compute costs.

For investors, the stakes are significant. Nvidia trades at a market cap of $5.49 trillion, and the stock has gained 17 percent year-to-date. The iShares Semiconductor ETF (SOXX) has surged 120 percent over the past 12 months. But if GPU prices fall as efficiency improves and utilization remains low, the collateral backing these new securities could deteriorate rapidly.

"When correlations start to deteriorate and borrower credit quality weakens simultaneously, risk accumulates quickly," White wrote. Nvidia would be forced to honor its insurance obligations while its own financing deals and core product values decline. Google and other companies using their balance sheets to support the capex boom face similar exposure.

Jane Sydenham, senior investment manager at Rathbones, echoed the concern: "The worry is that more and more money is going into these projects. Are they all going to earn the right return for the future?"

This article is for informational purposes only and does not constitute investment advice.